Genetic Predictors of Knee Pain in Persons With Mild to Moderate Osteoarthritis

Design

The current study examines the contribution of selected genetic variations upon knee OA pain using baseline clinical data collected as part of a double-blind, placebo-controlled randomized controlled trial (RCT) testing the effects of transcutaneous electrical nerve stimulation (TENS) on knee pain (Vance et al., 2012).

Sample and Setting

Participants who met the following conditions were enrolled: (a) a diagnosis of medial compartment knee OA; (b) age 18 to 60 years; (c) ability to ambulate to the mailbox and back; (d) a stable medication schedule over the prior 3 weeks; and (e) self-reported knee pain >3 on a 0 to 10 scale on the affected knee while weight bearing. Potential participants were not enrolled if they met one or more of the following criteria: (a) knee surgery within the past 6 months; (b) knee injection within the past 4 weeks; (c) presence of any serious medical condition; (d) uncontrolled diabetes mellitus or hypertension, dementia, or cognitive impairment; (e) permanent lower extremity sensory loss; and (f) prior TENS use. Upon enrollment, loss of sensation was further assessed through testing bilateral sharp dull pain recognition at the L3-S2 dermatomes as well as proprioception of the thumb and great toe, and participants with loss of sensation and lateral joint pain were excluded. Seventy-five participants were enrolled. The full CONSORT diagram can be found in the published RCT on the effect of TENS on knee OA pain (Vance et al., 2012). Approval for all recruitment and data collection procedures was obtained from the Human Subjects Institutional Review Boards at the investigators' institutions.

Variables and Instruments

Demographics. Demographic data were collected using an investigator-developed instrument, and included information about age, sex, race, marital status, education, duration of OA-related knee pain expressed in terms of number of months, and health status and history of surgical treatment excluding those related to knee OA. Health status variables consisted of yes/no responses to obesity, heart disease, hypertension, lung disease, diabetes, stomach disease, kidney disease, liver disease, cancer, anemia, depression, back pain, and rheumatoid arthritis.

Medication Use. For 34 of 75 participants, the names of current analgesic medications, including non-steroidal anti-inflammatory, narcotic, acetaminophen, and combination analgesic medications, and the time of last dose were known; 22 participants reported taking no medication. For the remaining 19 participants, medication data recorded was either unknown or incomplete. For the 34 individuals with known time and amount of last medication dose, the residual amounts of analgesic were calculated as acetaminophen equivalents assuming a decay model with half-lives based on published ranges, using the formula below.

D_exam refers to the amount of medication remaining at the time of exam (T_exam), D_last is the amount of the last dose taken at time T_last, and T_half is the half-life for that particular medication obtained from the hospital pharmacy online medication database, based on the product U.S. Food and Drug Administration–approved package inserts. Midpoints for half-lives with ranges were used in the data analysis. Residual quantities of analgesic were set to 0 for the 22 participants who reported using no medication.

Pain Sensitivity Assessment. An array of quantitative sensory tests was administered to each participant to measure pain for three types of stimuli in the following order: (a) cutaneous mechanical sensation and pain, (b) pressure pain, and (c) thermal pain. This approach to quantitative sensory testing has demonstrated adequate test–retest reliability in persons with knee OA (Wylde et al., 2011). The tests for each stimulus are described in greater detail in the following sections. Prior to administering the tests, three sites were marked on the participant 1 centimeter apart at the medial joint line (Kn) with an indelible marker. Three other sites were marked 1 inch apart on the anterior tibialis muscle (AT), with the top site marked 3 inches below the inferior border of the patella. Pain sensory tests were administered at each of these six sites, and each set of three measurements for the knee and AT were averaged into a single measurement, other than one notable exception. Due to the large size of the thermode stimulator probe, the thermal pain test was applied only to the middle point at the knee and AT. Tests were conducted on both the affected (ipsilateral) and unaffected (contralateral) knees; however, only pain measures on the affected knee were analyzed.

Cutaneous Mechanical Sensation and Pain (CMPT, six measures). Cutaneous mechanical testing was performed using von Frey filaments. Two sensation thresholds, SENS(Kn) and SENS(AT), and two pain thresholds, CMPT(Kn) and CMPT(AT), were measured by applying a series of 20 von Frey filaments in ascending order ranging from 0.08 to 1,813 g (0.08, 0.2, 0.36, 0.72, 1.73, 4.26, 5.87, 9.84, 12.37, 17.8, 35.02, 50.37, 66.28, 80.88, 112.41, 186.92, 447.04, 709, 1,228.27, and 1,813 g). Participants provided two self-report ratings on a 100-mm visual analogue scale (VAS), CMP-VAS(Kn) and CMP-VAS(AT), for a 6-g von Frey filament applied at each of the six sites on both legs. This method of assessing CMPT demonstrates adequate test–retest reliability (intraclass correlation coefficient [ICC] = 0.59) (Wylde et al. 2011).

Pressure Pain Threshold (PPT, two measures). A hand-held digital pressure algometer was used to measure pressure pain thresholds at the knee, PPT(Kn), and AT, PPT(AT), applying the 1 cm² circular probe and 40 kPa/s of pressure. Participants were instructed to press the hand-held response switch when the sensation first became painful. Prior to the actual measurement, participants trained themselves to respond with accuracy to the pressure applied on their non-dominant forearm. The actual measurement of PPT was then performed on the six marked sites on each knee and AT. PPTs measured in this way demonstrate high test–retest reliability (ICC = 0.83) (Wylde et al., 2011) as well as good interrater reliability with ICC scores ranging from 0.62 to 0.91 (Cheatham et al., 2018; Chesterton et al., 2007; O'Neill & O'Neill, 2015).

Thermal Pain Thresholds and Temporal Summation (HPT, HTS, four measures). Two heat pain threshold measures, HPT(Kn) and HPT(AT), and two heat temporal summation measures, HTS(Kn) and HTS(AT), were taken using a TSA 2 NeuroSensory Analyzer. This method of testing HPTs demonstrates moderate test–retest reliability (ICC = 0.77) (Wylde et al., 2011) and good to excellent interrater reliability (ICC = 0.52 to 0.86) (Moloney et al., 2011). For both threshold and temporal summation tests, the 5 cm² thermode was placed on the middle of the three marked test sites at the knee and AT, respectively. For HPT, the temperature was initially set to 37°C and increased to a maximum of 52°C. Participants indicated when they first felt pain (1/10 on a 0 to 10 scale, with higher scores indicating greater pain) by pressing a remote switch, which recorded the temperature and terminated the thermal stimulus. For HTS, a tonic heat stimulus of 45.5°C was applied for 20 seconds. After increasing to 45.5°C in the first 5 seconds and maintained for an additional 15 seconds, each participant rated pain on a VAS at 5-, 10-, and 15-second time points after the temperature of 45.5°C was reached. The average VAS score from three trials was used for each time point, and summed according to the formulae below:

Subjective Pain and Pain-Related Distress. In addition to the quantitative sensory assessment, measurement of the pain experience also included subjective measures of pain. These measures of pain included subjective pain intensity and pain-related distress at rest as well as subjective pain intensity and pain-related distress during movement.

At Rest (AtRest, two measures). Participants provided a self-report of resting pain intensity, AtRest(INT), by placing a mark on a horizontal 100-mm VAS, using the anchors of no pain (0) and worst imaginable pain (100) (Kahl & Cleland, 2005) as well as distress associated with the pain, AtRest(DIS) (Rodriguez-Fontenla et al., 2014), also on a 100-mm VAS and using the anchors no distress (0) and worst imaginable distress (100).

With Movement (Timed Up and Go [TUG], two measures). Subjective pain with movement was measured by asking participants to rate pain intensity, TUG(INT), and pain-related distress, TUG(DIS) (Rodriguez-Fontenla et al., 2014), on a VAS as described above following the TUG test. The TUG is a standardized, psychometrically sound measure of functional mobility (Podsiadlo & Richardson, 1991). On command, participants arise from a chair with no arm rests, ambulate 9.8 feet as quickly and safely as possible, turn, ambulate back, turn, and return to sitting in the chair. The walking distance is measured in advance and marked on the floor. Participants were timed from the point their upper back left the chair until the point they returned to a full sitting position with their back in contact with the chair. The times generated from the TUG test were not analyzed in this study.

Procedures

Recruitment. Active recruitment through flyers was used by the research team at the Orthopedic and Sports Medicine Department of a large Midwestern tertiary care center to collect study participants.

Data Collection. The first round of data collected at the examination comprised a demographic questionnaire, height and weight measurements, and the self-report pain and distress at rest. Quantitative sensory testing (QST) was performed, and finally participants completed the TUG test. All participants were administered QST by the same examiner using the same instrumentation, and consistently in the following order: (i) CMPT, (ii) PPT, (iii) HPT, and (iv) HTS. The testing order for each of the four areas (knee, AT belly, ipsilateral, and contralateral) and the three test sites were randomized to prevent an ordering effect of testing.

Sample Processing and Genotyping. Saliva samples were collected as a DNA source. DNA was extracted using Oragene^® DNA Isolation Kits. All genotypes were generated using Taqman SNP genotyping assays through Applied Biosystems with the exception of serotonin transporter (5HTT). Genotyping for the 5HTT promotor region insertion/deletion polymorphism was performed according to the methods described by Heils et al. (1996); polymerase chain reaction products were separated on a 2% agarose gel supplemented with Ethidium bromide and visualized by ultraviolet transillumination. Observed allele and genotype frequencies were consistent with those predicted by Hardy-Weinberg equilibrium. Seventy-four of 75 participants were successfully genotyped for DNA variants consisting of single nucleotide polymorphisms (SNPs) in several candidate genes. These genes were selected based on their known or hypothesized role in pain through either central or peripheral pain pathways or inflammatory pathways and included: nerve growth factor beta (NGFB), neurotrophic receptor tyrosine kinase 1 (NTRK1), endothelin 1 (EDN1), endothelin receptor A (EDNRA), endothelin receptor B (EDNRB), opioid receptor mu 1 (OPRM1), tachykinin precursor 1 (TAC1), tachykinin receptor 1 (TACR1), brain derived neurotrophic factor (BDNF), bradykinin receptor B1 (BDKRB1), 5HTT, interleukin-1 beta (IL1B), interleukin-6 (IL-6), estrogen receptor 2 (ESR2), and catechol-O-methyltransferase (COMT). Quality assurance through duplicate genotyping of samples resulted in a 100% concordance rate between the replicates.

Statistical Analysis

The statistical analysis procedure comprised: (a) synthesis of pain outcomes from pain measures, within and across instruments; (b) adjusting pain outcomes for demographic covariates; and (c) genetic association analysis of adjusted outcomes. Genetic association was assessed by analysis of each adjusted outcome individually as well as concurrent analysis of multiple outcomes. The step-by-step analysis procedure is outlined in Figure 1 and described in the following sections.

Figure 1. Flowchart for derivation of baseline osteoarthritis (OA) pain phenotypes. Note. SNP = single nucleotide polymorphism.

Creation of Pain Outcomes Within Test Instruments. Within each instrument, the original pain measures were examined for correlations, and, where appropriate, combined into composite outcomes by taking their averages if the pairwise correlation exceeded 80%. Figure 1 shows the correlation thresholds used, and Figure 2 shows pairwise correlation values. Correlations >80% were found within four pairs of pain measures; as a result, the 16 pain measures were reduced to 12. The eight original measures retained were PPT(Kn), PPT(AT), HPT(Kn), HPT(AT), SENS(Kn), SENS(AT), CMPT(Kn), and CMPT(AT). The four new averaged measures comprised: (a) HTS(Kn+AT): average of HTS(Kn) and HTS(AT); (b) CMP–VAS(Kn+AT): average of CMP–VAS(Kn) and CMP–VAS(AT); (c) AtRest(INT+DIS): average of AtRest(INT) and AtRest(DIS); and (d) TUG(INT+DIS): average of TUG(INT) and TUG(DIS). These 12 measures are labeled as Level 1 outcomes in Figure 1.

Figure 2. Results of correlation and exploratory factor analysis (EFA) on baseline pain measurements. Correlation within each instrument (A) and factors and loading values for cross-instrument EFA (B). Note. PPT = pressure pain threshold; Kn = medial joint line ; AT = anterior tibialis; HPT = heat pain threshold; HTS = heat temporal summation; SENS = sensation threshold; CMPT = cutaneous mechanical sensation and pain threshold; CMP = cutaneous mechanical pain; VAS = visual analogue scale; TUG = Timed Up and Go test; INT = intensity; DIS = distress; F = factor.

Composite Pain Outcomes Across Instruments. Exploratory factor analysis (EFA) was run on the 12 Level 1 outcomes to further use the correlations between them. Factor analysis is used to combine multiple variables into a smaller set of unobserved or latent variables that approximate the variability of the original outcomes within a margin of error. The new variables created are called factors. Factors are linear combinations of the original measures and can be used as new outcomes. EFA was performed using the minimum residual method (Harman & Jones, 1966), followed by parallel analysis (Humphreys & Montanelli, 1975), to determine the optimal number of factors. Parallel analysis identified four uncorrelated, independent factors labeled F1 through F4, which were then added to the set of Level 1 outcomes, producing a set of 16 Level 2 outcomes (Figure 1).

Adjustment for Demographic Variables. The 16 Level 2 outcomes were tested for correlation with gender, age, residual medication dose, obesity, back pain, and duration of OA-related knee pain using a multivariate linear regression framework. Residuals produced by the regression were taken as the final Level 3 pain outcomes (Figure 1) for genetic analysis. For example, gender and back pain were found to be significantly correlated to PPT(AT), which was then adjusted using the linear model shown below (βs denote regression coefficients).

Following adjustments for demographic variables, the resulting Level 3 outcomes were used as phenotypes within genetic association analysis.

Allele and Genotype Frequencies. None of the SNPs had excessively rare alleles, with the minimum allele frequency observed being 13%. Table A (available in the online version of this article) contains the allele and genotype frequencies for each SNP as observed in the study sample, along with the label of the minor allele, and the number of individuals for whom genotypes were available. The smallest p value for the Hardy-Weinberg equilibrium observed was for SNP rs1799971 within the OPRM1 gene on chromosome 6, with a p value close to 0.0001. High linkage dis-equilibrium was observed between rs5333 and rs10003447 in the EDNRA gene (r² = 0.96) and three SNPs (rs4633, rs4818, rs4680) within COMT (r² = 0.87 to 0.95).

Table A: Position, allele frequency, and genotype frequency of SNPs.

Single Outcome Genetic Association. Each of the 16 Level 3 outcomes (including those from the EFA) was tested against the 25 SNPs for genotype-specific effects using linear regression assuming an additive genetic model (Fingerlin et al., 2004) by coding each individual's geno-type at a SNP as the number of copies of its minor allele. All statistical analyses were performed using the R statistical language (R Core Team, 2017). Testing multiple genetic variants across the same set of participants raises the issue of multiple testing where the null hypothesis is rejected by chance more often than the significance threshold used. To correct for multiple testing, a Bonferroni adjusted p value of 0.2% was used instead of the customary 5% significance threshold.

Multi-Outcome Genetic Association. Pleiotropy, the phenomenon where one gene could be affecting more than one phenotype, is plausible in a complex phenotype such as pain. Concurrent association analysis of multiple pheno-types has greater power to detect a pleiotropic gene; therefore, each SNP was also analyzed for association to multiple pain outcomes using MultiPhen (O'Reilly et al., 2012) as an alternative to multivariate regression. In MultiPhen, the SNP genotype is modeled as a linear combination of the phenotypic outcomes in a reverse regression framework, using a proportional odds ordinal logistic model. The primary goal of this method is increased power to identify meaningful combinations of phenotypes, while making no distributional assumptions of the phenotypes. However, more sophisticated statistical modeling is necessary to refine the phenotype combinations, and/or assess genotypic effect on each phenotype thus tested. Therefore, only the significance of the association for multi-outcome regression is reported.

To conduct multi-outcome associations, Level 3 outcomes were grouped into the following five sets: (1) pressure pain: (PPT[Kn], PPT[AT]); (2) thermal pain: (HPT[Kn], HPT[AT], HTS[Kn+AT]); (3) mechanical pain (CMP): (SENS[Kn], SENS[AT], CMPT[Kn], CMPT[AT], CMP-VAS[Kn+AT]); (4) subjective pain (SRP): (AtRest[INT+DIS], TUG[INT+DIS]); and (5) combined pain comprising EFA factors (CPF): (F1, F2, F3, F4). In the analysis, the SNP genotype was coded as the number of minor alleles, to correspond to an additive genetic model, and interaction terms between pain outcomes within each group were excluded.

Characteristics of Phenotypic, Demographic, and Genetic Data

Distribution of Demographic Factors. The age of participants varied widely (range = 31 to 94 years), with a mean age of 56.3 years. A majority of participants were female (61%), White (90%), had greater than a high school education (57%), and were married (68%). Table 1 shows only the distribution of demographic factors that were tested for correlation to OA pain phenotypes. Table B (available in the online version of this article) contains the distribution of all quantitative, categorical, and binary variables among the 75 participants. Ranges, means, and standard deviations are shown for quantitative variables, including age, knee pain, and last medication dose, and counts are provided for the binary and categorical variables.

Table 1: Distribution of Demographic Variables Tested for Correlation With Osteoarthritis Pain Measures (N = 75)

Table B: Distribution of demographic variables

Factors Identified By Exploratory Factor Analysis. The optimal number of factors, as decided by parallel analysis, was set to four orthogonal factors. Figure 2 shows the relationship between the 12 measured Level 3 outcomes and the four factors, shown by arrows drawn from each factor to one or more observed pain outcomes. Factor loading values representing the magnitude of correlation between each factor and one or more observed outcomes are shown for each such relationship. For example, Factor 1 is correlated with AtRest(INT+DIS), TUG(INT+DIS), and cutaneous mechanical sensation tests. Factor 2 is correlated with cutaneous mechanical pain tests. Factor 3 is correlated with the pressure pain threshold, and Factor 4 is correlated with heat pain. With the exception of Factor 1, there is no evidence of shared underlying factors between the instruments based on correlation between the observed outcomes.

Correlation Between Demographic Factors and Level 2 Pain Outcomes. Several statistically significant relationships were identified between demographic variables and the Level 2 pain outcome variables. For example, gender was significantly correlated with pressure pain threshold at the knee and AT sites (PPT[Kn] and PPT[AT]); obesity was correlated with heat pain threshold (HPT[AT]) and pain at rest (AtRest[Kn+AT]); and back pain was correlated with pressure pain threshold (PPT[Kn]). Table 2 shows the summary characteristics of the 16 Level 3 pain outcomes after adjusting for demographic variables, as well as the correlation p values and effect sizes of correlated demographic variables. Note that Level 3 variables are residuals from linear regression models and therefore may have negative values.

Table 2: Distribution of Level 3 Pain Analysis Variables and Significantly Correlated Demographic Variables

Genetic Association Results

Single Outcome Genetic Association Results. An association p value exceeding the Bonferroni-adjusted significance threshold was observed for the adjusted cutaneous mechanical sensation (SENS[AT]) outcome and SNP rs6537485 in EDNRA (p = 0.0003; effect-size, β = 8.373 mg/cm²). The positive effect size indicates that the minor allele (A) of SNP rs6537485 was associated with higher thresholds for sensing cutaneous mechanical stimuli. SNP rs4818 in COMT was associated with thermal pain threshold (HPT[Kn]) (p = 0.004, β = −1.5°C). In this case, the minor allele (G) was associated with lower thresholds for heat pain. Association p values for the four pain factors did not reach the Bonferroni significance level. The smallest p value (0.01) was seen at SNP rs885845 in BDKRB1 for Factor 3 (pressure pain measures). Of note, however, is the observation that three SNPs (rs6269, rs4633, and rs4680) of five in COMT yielded p values smaller than the nominal 0.05 significance level in association with Factor 1 (self-report measures of pain and cutaneous sensation). Single outcome association p values are shown in the first 16 columns of Figure 3. The significance level of each association is represented as five intensity levels: (a) p ≥ 0.05 (nominal significance level); (b) 0.02 ≤ p < 0.05; (c) 0.005 ≤ p < 0.02; (d) 0.002 (Bonferroni significance level) ≤ p < 0.005; and (e) p < 0.002.

Figure 3. Genetic association p values for osteoarthritis (OA) pain phenotypes. Note. OA = osteoarthritis; SNP = single nucleotide polymorphism; NGFB = nerve growth factor beta; NTRK1 = neurotrophic receptor tyrosine kinase 1; IL1B = interleukin-1 beta; TACR1 = tachykinin receptor 1; EDNRA = endothelin receptor A; EDN1 = endothelin 1; OPRM1 = opioid receptor mu 1; IL-6 = interleukin-6; TAC1 = tachykinin precursor 1; BDNF = brain derived neurotrophic factor; EDNRB = endothelin receptor B; ESR2 = estrogen receptor 2; BDKRB1 = bradykinin receptor B1; 5HTT = serotonin transporter; COMT = catechol-O-methyltransferase; PPT = pressure pain threshold; HPT = heat pain threshold; HTS = heat temporal summation; Kn = medial joint line; AT = anterior tibialis; SENS = sensation threshold; CMPT = cutaneous mechanical sensation and pain threshold; CMP = cutaneous mechanical pain; VAS = visual analogue scale; INT = intensity; DIS = distress; F = factor.

Multi-Outcome Genetic Association Results. The multi-outcome analysis using the reverse regression framework did not yield any associations meeting the Bonferroni adjusted threshold. The last five columns of Figure 3 represent multi-outcome p values. The most significant association (p = 0.0077) was found for a polymorphism (C511T, rs16944) in IL1B and the combination of all pain factors (CPF) (Figure 3, final column). Table C (available in the online version of this article) lists p values for all SNPs.

Table C: Single phenotype and multiple phenotype association p-values of SNPs.